Anthropic just drew a hard line between production and deception. From Nov. 12, anyone using Claude — via the API or apps built on it — is barred from seeding search engines and AI systems with content that misrepresents where it really comes from.
The update, released Oct. 8, adds a new section called “Do Not Engage in Deceptive Campaigns or Artificial Activity.” Anthropic specifically calls out networks of sites posing as unaffiliated sources that “corroborate the same claims.”
What changes on Nov. 12
Earlier rules against fake accounts and fake news sites were scattered across election, fraud, privacy, and disinformation policies. The Sept. 15, 2025 version didn’t mention search engines or AI responses at all — which Anthropic says didn’t make the practice okay, just unnamed.
The new section is broader. It bans fake personas, fake outlets, and fake reviews, plus spreading content through “ostensibly independent outlets” to hide a shared source. Building tools for deceptive campaigns and selling those services as a product are prohibited too.
One big shift for publishers: automated publishing no longer appears on the high-risk use case list. The new list focuses on 11 areas such as legal, medical, and financial advice, and decisions about credit, housing, and employment. The rule about not submitting AI-assisted work without permission or attribution remains.
- Fake source networks: no posing as unrelated sites backing the same claim.
- Fake personas, outlets, and reviews: banned when they hide a shared origin.
- Deception as a service: building or selling manipulative tools is a violation.
The 70-site network behind the change
Anthropic tied the update to a September threat intelligence report. It found around 70 websites that looked like independent local news outlets but were connected to one France-based ad agency.
The scheme used Claude to create articles in a consistent format, each carrying three to four internal links, for automated publishing. Anthropic said the pieces were “specifically designed to boost” search authority. The network published at least 8,913 articles across about 20 languages, though most drew little real engagement.
Anthropic removed the Claude account and banned the organization. It didn’t report whether the sites actually improved rankings or showed up in AI-generated answers.
Why marketers should care
Google already classifies “attempting to manipulate generative AI responses in Google Search” as spam. Its May 15 documentation update applied that to AI answers in Search, and enforcement has been tough because Google polices websites while the content can come from anywhere.
Anthropic’s move hits the other end of the pipeline: the AI tool that generates the content. If a network’s authority rests on hidden ownership, it’s now a usage violation with possible account suspension, not just a ranking risk.
Your pre-Nov. 12 audit
Check any Claude-generated content you or an agency publish. Specifically look for content spread across a network of sites presented as independent. Then compare your review and disclosure process against the new high-risk list.
- Map every site in your network and how they link to each other.
- Add clear ownership disclosure or remove misleading pages.
- Document AI review steps for legal, medical, and financial content.
The policy will keep evolving. Anthropic updates its usage rules annually as Claude’s capabilities and risks change, so source transparency is now a workflow, not a one-time fix.
Source: Search Engine Journal
